Regime Shift Review

Detailing the Regime Shifts in Maine Coastal Current Behavior

{
  library(sf) 
  library(fvcom) 
  library(tidyverse)
  library(gmRi)
  library(patchwork)
  library(rnaturalearth)
  library(showtext)
  library(ncdf4)
  # Cyclic color palettes in scico
  # From: https://www.fabiocrameri.ch/colourmaps/
  library(scico)
  library(legendry)
  library(ggh4x)
  library(ecodata)
  library(trend)
}

# namespace conflicts
conflicted::conflict_prefer("select", "dplyr")
conflicted::conflict_prefer("filter", "dplyr")


# Set the theme
theme_set(theme_gmri_simple() +
    theme(strip.text.y = element_text(angle = 0),
          legend.position = "bottom", 
          legend.title.position = "top", 
          legend.title = element_text(hjust = 0.5), 
          strip.text = element_text(size = 8),
          axis.text = element_text(size = 8)))

# Project paths
lob_ecol_path <- cs_path("mills", "Projects/Lobster ECOL")
fvcom_path    <- cs_path("res", "FVCOM/Lobster-ECOL")
poly_paths    <- cs_path("mills", "Projects/Lobster ECOL/Spatial_Defs")


# Shapefiles
new_england <- ne_states("united states of america", returnclass = "sf") %>% 
  filter(postal %in% c("VT", "ME", "RI", "MA", "CT", "NH", "NY", "MD", "VA", "NJ", "DE", "NC", "PA", "WV"))
canada <- ne_states("canada", returnclass = "sf")

# # # Support functions for FVCOM
source(here::here("R/FVCOM_Support.R"))



# New areas factor levels
areas_northsouth <- c(
  "eastern maine",  
  "central maine",  
  "western maine",  
  "eastern mass",
  "southern mass",  
  "rhode island shore",
  "long island sound", 
  "new jersey shore",  
  "five fathom bank",  
  "virginia shore",
  "gom_gbk",         
  "sne")

# Use GMRI style
use_gmri_style_rmd()

# Stirnimann used these values in their paper:
# l = 5, 10, 15, 17.5 years, with monthly data
# Huber = 1
# Subsampling = (l + 1) / 3


# Load the function(s)
source(here::here("rstars-master","rSTARS.R"))

# Read Regions in
proj_path <- cs_path("mills", "Projects/Lobster ECOL")

# Load Shapefiles for inshore/offshore
poly_paths <- cs_path("mills", "Projects/Lobster ECOL/Spatial_Defs")



# NEW Areas

# clusters of statistical areas that align loosely with geography and management areas
inshore_areas <- read_sf(str_c(poly_paths,"spatial_defs_2025/12nm_poly_statarea_merge.shp")) %>% 
  janitor::clean_names() %>% 
  mutate(
    area_type = "nearshore-coastal",
    area_id = tolower(short_name))

# ecological production units
offshore_areas <- read_sf(str_c(poly_paths,"spatial_defs_2025/sne_gom_tocoast.shp"))  %>% 
  janitor::clean_names() %>% 
  mutate(
    area_type = "offshore-regional",
    area_id = tolower(region))

# Combine them
study_regions <- bind_rows(
  st_transform(dplyr::select(inshore_areas, area_id, geometry), st_crs(offshore_areas)), 
  dplyr::select(offshore_areas, area_id, geometry)
)

Reviewing STARS Regime Changes in Space/Time

This markdown reviews the various rstars regime shift results which were produced separately. I will begin at the largest geographic scales and work down to local timeseries:

Regime shifts for individual timeseries were tested using the STARS methodology. Any daily timeseries (temperature and salinity from ocean reanalysis models) were aggregated to a monthly temporal resolution, and any trends and seasonal cycles were removed.

The Marriott, Pope and Kendall (MPK) “pre-whitening” routine was used within the {rstars} algorithm to remove “red noise” (autoregressive processes, typically AR1) from the timeseries.

For more details on trend removal and pre-whitening methods see Rodionov 2006.

Shelf-Scale Ecodata Indices

There are 2-3 climate and oceanographic timeseries that operate at the broad regional scale of the Northeast shelf. These are:

1. The Gulf Stream Index (a metric indicating the North/South position of the Gulf Stream) based on SSH
2. The Northeast Channel Slopewater Proportions (the percentage of various water masses at the 150-200m depth entering GOM, using NERACOOS buoy data)
3. The North Atlantic Oscillation (atmospheric pressure differential between icelandic low and the Azores High)

These three indices affect conditions over large spatial scales and and are likely to either directly or indirectly impact more local scale changes.

These three metrics are available in the ecodata r package and can be pulled directly from the package.

# GSI
# Why is there more than one value per month?
gsi <- ecodata::gsi %>% #glimpse()
  mutate(Time = as.Date(str_c(str_replace(Time, "[.]", "-"), "-01")))


# use the old one
gsi_old <- ecodata::gsi_old %>% #glimpse()
  mutate(Var = "gulf stream index old") %>% 
  mutate(Time = as.Date(str_c(str_replace(Time, "[.]", "-"), "-01")))


# NAO
nao <- ecodata::nao%>% 
  mutate(Time = as.Date(str_c(Time, "-01-01")))

# Put them together to plot
shelf_indices <- bind_rows(list(gsi, gsi_old, nao))


# Plot the residuals from the trend
ggplot(shelf_indices, aes(Time, Value)) + 
  geom_line(linewidth = 0.6, alpha = 0.8) + 
  facet_grid(Var ~ ., scales = "free", labeller = label_wrap_gen(width= 8)) +
  guides(color = guide_legend(nrow = 2)) +
  labs(title = "Shelf Scale Ocean/Climate Metrics - Raw")

The Gulf Stream indices come as two monthly datasets, the other indices are annual.

Trends in Ecodata Indicators

A Mann-Kendall test can be used to determine whether there is any monotonic (increasing/decreasing) trends in a time series. I will be checking each timeseries along the way using this method to keep a tab on whether or not long-term trends existed, which may be relevant to ecosystem change regardless of stars regime results.

# Perform test of monotonic trends and report as table
shelf_indices %>% 
  split(.$Var) %>% 
  map_dfr(function(x){
    # check pacing
    x_2000 <- filter(x, year(Time) == 2000)
    tempo <- if_else(nrow(x_2000) == 12, 12, 1)
    message(str_c(x$Var[[1]], " has ", tempo, " values"))
    
    trend <- trend::mk.test(x$Value)$p.value < 0.05
    rate <- ifelse(
      trend, 
      coef(lm(Value ~ Time, data = x))[[2]] * tempo * 10,
      NA)
   
    return(tibble(
      "Trend?" = trend,
      "Decadal Rate" = round(rate, 3)))
  }, .id = "Var") %>% 
  gt::gt() %>% 
  gt::tab_header(
    title = "Ecodata Shelf-Scale Long Term Trends",
    subtitle = "Monotonic Trends Evaluated by Mann-Kendall Test")  %>%
  gt::fmt_missing(columns = everything(), missing_text = "")

Ecodata Shelf-Scale Long Term Trends
Monotonic Trends Evaluated by Mann-Kendall Test
Var	Trend?	Decadal Rate
gulf stream index	TRUE	0.009
gulf stream index old	FALSE
north atlantic oscillation	FALSE
western gulf stream index	TRUE	0.008

The Gulf Stream and Western Gulf Stream indices show a long term increasing trend, with some evidence in the GSI that this quickly changed course around 2023.

shelf_indices %>% 
  filter(Var == "gulf stream index", year(Time) > 2010) %>% 
  ggplot(aes(Time, Value)) +
  geom_line() +
  geom_vline(xintercept = as.Date("2023-01-01"), linetype = 3, color = "gray50") +
  facet_grid(Var~.) + 
  labs(title = "Recent Course-Change in Gulf Stream Index",
       subtitle = "Rapid Gulf Stream Position Reset After 2023")

Because the presence of long-term trends can influence the results of tests for breakpoint/mean shifts, any long-term trends for each metric have been removed prior to regime shift tests on these metrics.

# Run the shift test for summertime PCA
shelf_indices_detrended <- shelf_indices %>%
  split(.$Var) %>%
  map_dfr(function(.x){

    # Detrend
    .x <- .x %>% 
      arrange(Time) %>% 
      mutate(
        time = Time,
        yr_num = year(time))
    
    # annual trend
    trend_mod <- lm(Value ~ Time, data = .x)
    
    # save the results
    .x <- broom::augment(x = trend_mod) %>%
      rename(
        trend_fit = .fitted,
        trend_resid = .resid) %>% 
      full_join(.x, join_by(Time, Value)) %>% 
      mutate(trend_resid = if_else(is.na(Value), NA, trend_resid))
    
    return(.x)})

The following plot shows what these timeseries look like after the removal of any monotonic trend:

# Plot the residuals from the trend
ggplot(shelf_indices_detrended, aes(Time, trend_resid)) + 
  geom_line(linewidth = 0.6, alpha = 0.8) + 
  facet_grid(Var ~ ., scales = "free", labeller = label_wrap_gen(width= 8)) +
  guides(color = guide_legend(nrow = 2)) +
  labs(title = "Shelf Scale Ocean/Climate Metrics - Detrended")

Bacause the slopewater proportion contains NA values, we cannot evaluate it for breaks unless we impute missing values somehow or take a subset of time that is uninterrupted.

# Run the regime shift test
shelf_indices_rstars <- shelf_indices_detrended %>%
  split(.$Var) %>%
  map_dfr(function(.x){
    
    # Set the cutoff length, change it based on monthly/annual
    cutoff_length <- ifelse(
      str_detect(.x$Var[[1]], "index"),
      12*7,
      7)
    
    # This is only here because we have duplicate dates in the GSI
    .x <- distinct(.x, Time, .keep_all = T)
    
    # Get the results from that
    x_rstars <- rstars(
      data.timeseries = as.data.frame(
        .x[,c("Time", "trend_resid")]),
      l.cutoff = cutoff_length,
      pValue = 0.05,
      Huber = 1,
      Endfunction = T,
      preWhitening = T,
      OLS = F,
      MPK = T,
      IP4 = F,
      SubsampleSize = (cutoff_length + 1)/3,
      returnResults = T) %>% 
      mutate(
        EPU = .x$EPU[[1]],
        Value = .x$Value,
        shift_direction = case_when(
          RSI > 0 ~ "Shift Up", 
          RSI < 0 ~ "Shift Down",
          TRUE ~ NA))
    },
    
    .id = "Var"
  )

The results can be seen below:

# Summarise the breakpoint locations
shelf_shift_points <- shelf_indices_rstars %>% 
  filter(RSI != 0) %>% 
  dplyr::select(Time, Var, EPU, shift_direction)


# Plot the breaks over the monthly data
ggplot() +
  # Add a vertical line using geom_segment with an arrow
  geom_segment(
    data = filter(
      shelf_shift_points, 
      str_detect(shift_direction, "Up")),
    linewidth = 0.8,
    aes(x = Time, xend = Time, y = -Inf, , yend = Inf, 
        color = shift_direction),  
    arrow = arrow(length = unit(0.3, "cm"), ends = "last")) +
  geom_segment(
    data = filter(
      shelf_shift_points, 
      !str_detect(shift_direction, "Up")),
    linewidth = 0.8,
    aes(x = Time, xend = Time, y = -Inf, , yend = Inf, 
        color = shift_direction),  
    arrow = arrow(length = unit(0.3, "cm"), ends = "first")) +
  geom_line(
    data = shelf_indices_rstars,
    aes(Time, trend_resid),
    linewidth = 0.4, alpha = 0.5) +
  scale_color_gmri() +
  facet_grid(EPU * Var~., labeller = label_wrap_gen(width = 8)) +
  scale_x_date(
    breaks = seq.Date(
      from = as.Date("1950-01-01"),
      to = as.Date("2020-01-01") ,
      by = "10 year"),
     limits = as.Date(c("1970-01-01", "2020-01-01")),
    labels = scales::label_date_short()) +
  theme(
    legend.position = "bottom",
    strip.text.y = element_text(angle = 0)) + 
  labs(color = "",
       y = "Measurement",
       x = "Date",
       title = "Shelf-Scale Ocean/Climate Metrics - STARS Changepoints",
       subtitle = "Performed on Detrended Timeseries")

Based on these results, there is evidence for breakpoints in the Gulf Stream Indices, and not in the NAO index.

shelf_shift_points %>% 
  group_by(Var) %>% 
  arrange(Time, EPU) %>% 
  gt::gt() %>% 
  gt::tab_header(title = "Shelf-Scale Breaks")

Shelf-Scale Breaks
Time	EPU	shift_direction
gulf stream index
1957-12-01	All	Shift Down
1971-09-01	All	Shift Up
1977-08-01	All	Shift Down
2003-01-01	All	Shift Down
2012-02-01	All	Shift Up
western gulf stream index
1962-02-01	All	Shift Down
1971-07-01	All	Shift Up
2003-01-01	All	Shift Down
2011-12-01	All	Shift Up
gulf stream index old
2003-01-01	All	Shift Down

EPU-Scale Ecodata Indices

For the EPU-scale metrics we have a number of indices available from the ecodata package:

1. The cold-pool index
2. The Northeast Channel Slopewater Proportion (from NERACOOS Buoy N)
3. Metrics of primary production and zooplankton community
4. Temperature and salinity timeseries specific to each area

Temperature and salinity is from either GLORYS or FVCOM, primary productivity is satellite derived (OC-CCI, SeaWiFS, MODIS-Aqua), and the zooplankton community indices are from the Gulf of Maine CPR transect.

# # There are a ton here. We want primary productivity / chlor a, and maybe anomalies
# chl_pp <- ecodata::chl_pp %>% 
#   filter(str_detect(Var, "MONTHLY")) %>% 
#   filter(str_detect(Var, "PPD|CHLOR_A")) %>% 
#   separate(col = "Time", into = c("Period", "Time"), sep = "_") %>% 
#   mutate(Time = as.Date(
#     str_c(
#       str_sub(Time, 1, 4),
#       str_sub(Time, 5, 6), 
#       "01",
#       sep = "-")))

# Annual will make life easier
annual_chl_pp <- ecodata::annual_chl_pp %>% 
  filter(str_detect(Var, "MEAN")) %>% 
  separate(col = "Time", into = c("Period", "Time"), sep = "_") %>% 
  mutate(Time = as.Date(
    str_c(
      Time,
      "01-01",
      sep = "-")))


# Just take one cold pool index for now
cold_pool <- ecodata::cold_pool %>% #distinct(Source)
  filter(Var == "cold_pool_index") %>% 
  mutate(
    Var = str_c(Source, Var, sep = "_"),
    Time = as.Date(
    str_c(
      Time,
      "01-01",
      sep = "-")))

# Slopewater
slopewater <- ecodata::slopewater %>% 
  mutate(Time = as.Date(str_c(Time, "-01-01")))  %>% 
  filter(Time > as.Date("1990-01-01")) %>% 
  drop_na()


# Combine those
epu_indices <- bind_rows(annual_chl_pp, slopewater, cold_pool) %>% 
  group_by(Var, EPU) %>% 
  arrange(Time) 


# Plot them
ggplot(epu_indices, aes(Time, Value)) +
  geom_line(linewidth = 0.6, alpha = 0.8) +
  facet_grid(Var * EPU ~ ., scales = "free", labeller = label_wrap_gen(width= 8)) +
  guides(color = guide_legend(nrow = 3)) +
  scale_x_date(
    breaks = seq.Date(
      from = as.Date("1950-01-01"),
      to = as.Date("2020-01-01") ,
      by = "10 year"),
     limits = as.Date(c("1970-01-01", "2020-01-01")),
    labels = scales::label_date_short()) +
  guides(color = guide_legend(nrow = 2)) +
  labs(title = "EPU Scale Ocean/Ecological Metrics - Raw")

Long-Term EPU Scale Trends

# Perform test of monotonic trends and report as table
epu_indices %>% 
  mutate(var_epu = str_c(Var, EPU, sep = "X")) %>% 
  split(.$var_epu) %>% 
  map_dfr(function(x){
    
    
    # check pacing
    # Set the cutoff length, change it based on monthly/annual
    cutoff_length <- 7 # 7 years for annual
    trend <- trend::mk.test(x$Value)$p.value < 0.05
    rate <- ifelse(
      trend, 
      coef(lm(Value ~ Time, data = x))[[2]] * 12 * 10,
      NA)
   
    return(tibble(
      "Trend?" = trend,
      "Decadal Rate" = round(rate, 3)))
    }, 
  .id = "var_epu") %>% 
  separate(var_epu, into = c("Var", "EPU"), sep = "X") %>% 
  gt::gt() %>% 
  gt::tab_header(
    title = "Ecodata Shelf-Scale Long Term Trends",
    subtitle = "Monotonic Trends Evaluated by Mann-Kendall Test") %>%
  gt::fmt_missing(columns = everything(), missing_text = "")

Ecodata Shelf-Scale Long Term Trends
Monotonic Trends Evaluated by Mann-Kendall Test
Var	EPU	Trend?	Decadal Rate
CHLOR_A_ANNUAL_MEAN	GB	TRUE	0.002
CHLOR_A_ANNUAL_MEAN	GOM	TRUE	0.004
CHLOR_A_ANNUAL_MEAN	MAB	FALSE
GLORYS_cold_pool_index	MAB	TRUE	0.016
LSLW proportion ne channel	GOM	FALSE
MOM6_cold_pool_index	MAB	FALSE
PPD_ANNUAL_MEAN	GB	TRUE	0.002
PPD_ANNUAL_MEAN	GOM	TRUE	0.002
PPD_ANNUAL_MEAN	MAB	TRUE	0.001
ROMS_cold_pool_index	MAB	TRUE	0.013
WSW proportion ne channel	GOM	FALSE

Detrending Ecodata Indicators

Most of these indicators have data at the monthly time-scale. To aid in regime change detection long-term year over year changes have been removed.

# Detrend the epu stuff

epu_indices_detrended <- epu_indices %>% 
  mutate(var_epu = str_c(Var, EPU, sep = "X")) %>% 
  split(.$var_epu) %>%
  map_dfr(function(.x){

    # Detrend
    .x <- .x %>% 
      arrange(Time) %>% 
      mutate(
        time = Time,
        yr_num = year(time))
    
    # annual trend
    trend_mod <- lm(Value ~ Time, data = .x)
    
    # save the results
    .x <- broom::augment(x = trend_mod) %>%
      rename(
        trend_fit = .fitted,
        trend_resid = .resid) %>% 
      full_join(.x, join_by(Time, Value)) %>% 
      mutate(trend_resid = if_else(is.na(Value), NA, trend_resid))
    
    return(.x)})


# Plot detrended
ggplot(epu_indices_detrended, aes(Time, trend_resid)) + 
  geom_line(linewidth = 0.6, alpha = 0.8) + 
  facet_grid(EPU * Var ~ ., scales = "free", labeller = label_wrap_gen(width= 8)) +
  guides(color = guide_legend(nrow = 3)) +
    scale_x_date(
    breaks = seq.Date(
      from = as.Date("1950-01-01"),
      to = as.Date("2020-01-01") ,
      by = "10 year"),
     limits = as.Date(c("1970-01-01", "2020-01-01")),
    labels = scales::label_date_short()) +
  labs(title = "EPU Scale Ocean/Climate Metrics - Detrended",
       y = "Metric")

Once detrended, they can be checked for signs of regime changes.

# Run rstars for those, temperature and salinity are done


# Run the regime shift test
epu_indices_rstars <- epu_indices_detrended %>%
  #filter(str_detect(Var, "proportion") == FALSE) %>% 
  split(.$var_epu) %>%
  map_dfr(function(.x){
    
    # This is only here because we have duplicate dates in the GSI
    .x <- distinct(.x, Time, .keep_all = T)
    
    # cutoff length
    cutoff_length <- 7
    
    # Get the results from that
    x_rstars <- rstars(
      data.timeseries = as.data.frame(
        .x[,c("Time", "trend_resid")]),
      l.cutoff = cutoff_length,
      pValue = 0.05,
      Huber = 1,
      Endfunction = T,
      preWhitening = T,
      OLS = F,
      MPK = T,
      IP4 = F,
      SubsampleSize = (cutoff_length + 1)/3,
      returnResults = T) %>% 
      mutate(
        Value = .x$Value,
        shift_direction = case_when(
          RSI > 0 ~ "Shift Up", 
          RSI < 0 ~ "Shift Down",
          TRUE ~ NA))
    
    },
    
    .id = "var_epu"
  ) %>% 
  separate(var_epu, into = c("Var", "EPU"), sep = "X")

Primary Production and Cold-Pool Dynamics

The results from the STARS algorithm can be seen below:

# Summarise the breakpoint locations
epu_shift_points <- epu_indices_rstars %>% 
  filter(RSI != 0) %>% 
  dplyr::select(Time, Var, EPU, shift_direction)


# Plot the breaks over the monthly data
ggplot() +
   geom_vline(
    data = epu_shift_points,
    aes(
      xintercept = Time,
      color = shift_direction),
    linewidth = 1.5) +
  geom_line(
    data = epu_indices_rstars,
     aes(Time, Value),
    linewidth = 0.4, alpha = 0.5) +
  scale_color_gmri() +
  facet_grid(EPU * Var~., labeller = label_wrap_gen(width = 8), scales = "free") +
  scale_x_date(
    breaks = seq.Date(
      from = as.Date("1950-01-01"),
      to = as.Date("2020-01-01") ,
      by = "10 year"),
     limits = as.Date(c("1970-01-01", "2020-01-01")),
    labels = scales::label_date_short()) +
  theme(
    legend.position = "bottom",
    strip.text.y = element_text(angle = 0)) + 
  labs(color = "",
       y = "Measurement",
       x = "Date",
       title = "EPU Scale Ocean/Climate Metrics - Detrended",
       subtitle = "STARS Regime Shifts")

Based on these results, we see no breakpoints in EPU-Scale measures of primary production or the cold pool dynamics.

epu_shift_points %>% 
  group_by(Var) %>% 
  arrange(Time, EPU) %>% 
  gt::gt() %>% 
  gt::tab_header(title = "Ecodata EPU Scale Breakpoints")

Ecodata EPU Scale Breakpoints
Time	EPU	shift_direction

EPU-Scale Temperature and Salinity

Temperature and salinity timeseries from FVCOM were processed and had their STARS testing done separately in STARS_FVCOM.qmd. here are their results:

# These are the raw monthly timeseries:
# write_csv(surf_temp_monthly_shifts_raw, here::here("rstars_results/lobecol_stemp_monthly_shifts_raw.csv"))
# write_csv(bot_temp_monthly_shifts_raw, here::here("rstars_results/lobecol_btemp_monthly_shifts_raw.csv"))

# Load the data for the new regions
daily_fvcom_temps <- read_csv(
  str_c(lob_ecol_path, "FVCOM_processed/area_timeseries/new_regions_fvcom_temperatures_daily.csv")
) %>% 
  mutate(
    yday = lubridate::yday(time),
    year = lubridate::year(time),
    month = lubridate::month(time),
    depth_type = if_else(area_id %in% c("gom_gbk", "sne"), "offshore", "nearshore"),
    area_id = factor(area_id, levels = areas_northsouth))




# Run the monthly versions
monthly_fvcom_temps <- daily_fvcom_temps  %>%
  group_by(area_id, year, month, depth_type) %>%
  summarise(
    surface_t = mean(surface_t, na.rm = T),
    bottom_t  = mean(bottom_t, na.rm = T),
    .groups = "drop") %>%
  mutate(
    yr_num = as.numeric(as.character(year)),
    month = factor(month),
    time = as.Date(
      str_c(
        year,
        str_pad(month, side = "left", pad = "0", width = 2),
        "15", sep = "-"))) %>% 
  rename(
    surface_temperature = surface_t,
    bottom_temperature = bottom_t) %>% 
  pivot_longer(ends_with("temperature"), names_to = "var", values_to = "values")


# Monthly Salinity
monthly_fvcom_sal <- read_csv(
  str_c(lob_ecol_path, "FVCOM_processed/area_timeseries/new_regions_fvcom_salinity_monthly_gom3.csv")) %>% 
  mutate(
    yr_num = as.numeric(as.character(year(time))),
    month = factor(lubridate::month(time)),
    depth_type = if_else(area_id %in% c("gom_gbk", "sne"), "offshore", "nearshore"),
    area_id = factor(area_id, levels = areas_northsouth)
  ) %>% 
  pivot_longer(ends_with("salinity"), names_to = "var", values_to = "values")

# Combine these
fvcom_tempsal_raw <- bind_rows(
  monthly_fvcom_temps, monthly_fvcom_sal
)

Temp/Sal Trends

# Perform test of monotonic trends and report as table
fvcom_trends <- fvcom_tempsal_raw  %>% 
  mutate(var_area = str_c(var, area_id, sep = "X")) %>% 
  split(.$var_area) %>% 
  map_dfr(function(x){
    
    
    # check pacing
    # Set the cutoff length, change it based on monthly/annual
    cutoff_length <- 7 # 7 years for annual
    trend <- trend::mk.test(x$values)$p.value < 0.05
    rate <- ifelse(
      trend, 
      coef(lm(values ~ time, data = x))[[2]] * 12 * 10,
      NA)
   
    return(tibble(
      "Trend?" = trend,
      "Decadal Rate" = round(rate, 3)))
    }, 
  .id = "var_area") %>% 
  separate(var_area, into = c("var", "area_id"), sep = "X") 



fvcom_trends %>% 
  filter(#str_detect(var, "temperature"),
         `Trend?`,
         area_id %in% c("gom_gbk", "sne")) %>% 
  group_by(area_id) %>% 
  gt::gt() %>% 
  gt::tab_header(
    title = "FVCOM Offshore Long Term Temperature Trends",
    subtitle = "Monotonic Trends Evaluated by Mann-Kendall Test") %>%
  gt::fmt_missing(columns = everything(), missing_text = "")

FVCOM Offshore Long Term Temperature Trends
Monotonic Trends Evaluated by Mann-Kendall Test
var	Trend?	Decadal Rate
sne
bottom_salinity	TRUE	0.002
bottom_temperature	TRUE	0.010
surface_temperature	TRUE	0.014
gom_gbk
bottom_temperature	TRUE	0.009
surface_salinity	TRUE	-0.001
surface_temperature	TRUE	0.013

Temp/Sal STARS Breaks

In STARS_FVCOM.qmd I used temperature timeseries to explore the impacts of performing regime shift testing on raw or detrended monthly data. These tests helped reinforce that the suggested preprocessing (detrending etc.) did help isolate step-changes in the timeseries which may be related to a regime change.

The following breaks were identified from these detrended series:

# Load the regime shift results from STARS_FVCOM.qmd
# These are from detrended data
surf_sal_monthly_shifts <- read_csv(
  here::here("rstars_results/lobecol_ssal_monthly_shifts_detrended.csv")) %>% 
  mutate(Var = "Surface Salinity") %>% 
  rename(detrended_vals = ssal_model_resid) 
bot_sal_monthly_shifts <- read_csv(
  here::here("rstars_results/lobecol_bsal_monthly_shifts_detrended.csv"))  %>% 
  mutate(Var = "Bottom Salinity") %>% 
  rename(detrended_vals = bsal_model_resid)
surf_temp_monthly_shifts <- read_csv(
  here::here("rstars_results/lobecol_stemp_monthly_shifts_detrended.csv"))  %>% 
  mutate(Var = "Surface Temperature") %>% 
  rename(detrended_vals = stemp_model_resid)
bot_temp_monthly_shifts <- read_csv(
  here::here("rstars_results/lobecol_btemp_monthly_shifts_detrended.csv"))  %>% 
  mutate(Var = "Bottom Temperature") %>% 
  rename(detrended_vals = btemp_model_resid)

# Put them together
tempsal <- bind_rows(
  list(surf_sal_monthly_shifts, bot_sal_monthly_shifts, surf_temp_monthly_shifts, bot_temp_monthly_shifts)) %>%
  mutate(shift_direction = case_when(
          RSI > 0 ~ "Shift Up", 
          RSI < 0 ~ "Shift Down",
          TRUE ~ NA),
         area_id = factor(area_id, levels = areas_northsouth))
offshore_tempsal <- tempsal %>% filter(area_id %in% tolower(c("GOM_GBK", "SNE")))
inshore_tempsal <- tempsal %>% filter(area_id %in% tolower(c("GOM_GBK", "SNE")) == FALSE)


# Pull the shift points
tempsal_shifts <- tempsal %>% filter(RSI != 0)


# Split surface and bottom
offshore_tempsal_shifts <- offshore_tempsal %>% 
  filter(RSI != 0) %>% 
  dplyr::select(time, Var, area_id, shift_direction)
inshore_tempsal_shifts <- inshore_tempsal %>% 
  filter(RSI != 0) %>% 
  dplyr::select(time, Var, area_id, shift_direction)

# Plot the offshore shifts


# Plot the breaks over the monthly data
ggplot() +
   # Add a vertical line using geom_segment with an arrow
  geom_segment(
    data = filter(
      offshore_tempsal_shifts, 
      str_detect(shift_direction, "Up")),
    linewidth = 0.8,
    aes(x = time, xend = time, y = -Inf, yend = Inf, 
        color = shift_direction),  
    arrow = arrow(length = unit(0.3, "cm"), ends = "last")) +
  geom_segment(
    data = filter(
      offshore_tempsal_shifts, 
      !str_detect(shift_direction, "Up")),
    linewidth = 0.8,
    aes(x = time, xend = time, y = -Inf, yend = Inf, 
        color = shift_direction),  
    arrow = arrow(length = unit(0.3, "cm"), 
                  ends = "first")) +
  geom_line(
    data = offshore_tempsal,
     aes(time, detrended_vals),
    linewidth = 0.4, alpha = 0.5) +
  scale_color_gmri() +
  facet_nested(area_id * Var~., labeller = label_wrap_gen(width = 8), scales = "free") +
  scale_x_date(
    breaks = seq.Date(
      from = as.Date("1950-01-01"),
      to = as.Date("2020-01-01") ,
      by = "10 year"),
     limits = as.Date(c("1978-01-01", "2020-01-01")),
    labels = scales::label_date_short()) +
  theme(
    legend.position = "bottom",
    strip.text.y = element_text(angle = 0)) + 
  labs(color = "",
       y = "Measurement",
       x = "Date",
       title = "FVCOM EPU Scale Temp/Sal Metrics",
       subtitle = "STARS Regime Shifts")

A change in SNE salinity appears to have occured around 1992.

Surface temperatures fell in SNE around 2002, but they rose again in 2011 along with GOM+GBK the same year.

offshore_tempsal_shifts %>% 
  group_by(Var) %>% 
  arrange(time, area_id) %>% 
  gt::gt() %>% 
  gt::tab_header(title = "EPU Scale Temp+Sal Breaks")

EPU Scale Temp+Sal Breaks
time	area_id	shift_direction
Surface Temperature
2002-11-15	gom_gbk	Shift Down
2002-11-15	sne	Shift Down
2009-11-15	gom_gbk	Shift Up
2011-05-15	sne	Shift Up

CPR Community PCA Index

Work by Andy Pershing helped develop an understanding that the Gulf of Mane’s zooplankton community in a given year is often one of two groups with different life history and size characteristics. There is a large copepod community, of which Calanus finmarchicus (a large bodied, lipid rich species) is prominent, and a second community which is composed of smaller-bodied and more opportunistic zooplankton species. These two communities compete for the same prey resources, and are typically out of phase with one-another. A principal component analysis using the continuous plankton recorded data has been used as a proxy for which community is dominant each year.

# From Pershing & Kemberling
# PC1 explains 53.62% of variance
# PC2 explains 27.9%
# PC1 is associated with centropages, oithona, para-pseudocalanus
# PC2 is C. Fin, Metridia, & Euphausiacea

# Load the CPR data
cpr_community <- read_csv(here::here("local_data", "cpr_focal_pca_timeseries_period_1961-2017.csv")) %>% 
  rename(
    PC1_small_zoo = `First Mode`,
    PC2_large_zoo = `Second Mode`) %>% 
  select(-c(pca_period, taxa_used)) %>% 
  pivot_longer(cols = starts_with("PC"), names_to = "Var", values_to = "value")

Taking PCA timeseries as proxies for those communities and evaluating them for breakpoints gives the following results.

# Run breakpoints in CPR PCA


# Run the regime shift test
cpr_indices_rstars <- cpr_community %>%
  filter(year > 1976) %>% 
  mutate(EPU = "GOM",
         var_epu = str_c(Var, "X", EPU)) %>% 
  split(.$var_epu) %>%
  map_dfr(function(.x){
    
    # detrend
    trend_mod <- lm(value ~ year, data = .x)
    .x$trend_resid <- resid(trend_mod)
    
    # cutoff length
    cutoff_length <- 7
    
    # Get the results from that
    x_rstars <- rstars(
      data.timeseries = as.data.frame(
        .x[,c("year", "trend_resid")]),
      l.cutoff = cutoff_length,
      pValue = 0.05,
      Huber = 1,
      Endfunction = T,
      preWhitening = T,
      OLS = F,
      MPK = T,
      IP4 = F,
      SubsampleSize = (cutoff_length + 1)/3,
      returnResults = T) %>% 
      mutate(
        Value = .x$value,
        shift_direction = case_when(
          RSI > 0 ~ "Shift Up", 
          RSI < 0 ~ "Shift Down",
          TRUE ~ NA))
    
    },
    
    .id = "var_epu"
  ) %>% 
  separate(var_epu, into = c("Var", "EPU"), sep = "X") %>% 
  mutate(time = as.Date(str_c(year, "-01-01")))

# Summarise the breakpoint locations
cpr_shift_points <- cpr_indices_rstars %>% 
  filter(RSI != 0) %>% 
  dplyr::select(time, Var, EPU, shift_direction)


# Plot the breaks over the monthly data
ggplot() +
   geom_vline(
    data = cpr_shift_points,
    aes(
      xintercept = time,
      color = shift_direction),
    linewidth = 1.5) +
  geom_line(
    data = cpr_indices_rstars,
     aes(time, trend_resid),
    linewidth = 0.4, alpha = 0.5) +
  scale_color_gmri() +
  facet_grid(Var ~ EPU, labeller = label_wrap_gen(width = 8), scales = "free") +
  scale_x_date(
    breaks = seq.Date(
      from = as.Date("1950-01-01"),
      to = as.Date("2020-01-01") ,
      by = "10 year"),
     limits = as.Date(c("1970-01-01", "2020-01-01")),
    labels = scales::label_date_short()) +
  theme(
    legend.position = "bottom",
    strip.text.y = element_text(angle = 0)) + 
  labs(color = "",
       y = "Measurement",
       x = "Date",
       title = "CPR Zooplankton Community Metrics",
       subtitle = "STARS Regime Shifts")

ECOMON Community PCA Index

# Abundance per 100m3 for different taxa
# ecodata::zoo_regime %>% distinct(Var) %>% pull() %>% sort()
ecomon_zoo <- ecodata::zoo_regime
ecomon_zoo %>% 
  filter(!str_detect(Var, "fish|clauso|gas")) %>% 
  ggplot(aes(Time, Value)) +
  geom_line() +
  facet_grid(Var ~ EPU)

# We might be able to just pull out the seven focal species and then repeat the PCA process...
# Issues:
# ecomon doesn't split the calanus into adult/juvenile
# Para and Pseaudocalana are split
# Euph also has a Euph1

Large and Small Copepod Index

https://noaa-edab.github.io/tech-doc/zoo_abundance_anom.html?q=zoopl#copepod

Abundance anomalies are computed from the expected abundance on the day of sample collection. Abundance anomaly time series are constructed for Centropages typicus, Pseudocalanus spp., Calanus finmarchicus, and total zooplankton biovolume. The small-large copepod size index is computed by averaging the individual abundance anomalies of Pseudocalanus spp., Centropages hamatus, Centropages typicus, and Temora longicornis, and subtracting the abundance anomaly of Calanus finmarchicus. This index tracks the overall dominance of the small bodied copepods relative to the largest copepod in the Northeast U.S. region, Calanus finmarchicus.

# This has "LgCopepods" & "SmCopepods", which could produce large/small index
# ecodata::zoo_abundance_anom %>% distinct(Var) %>% pull() %>% sort()


zoo_lg_small <- ecodata::zoo_abundance_anom %>% 
  filter(Var %in% c("LgCopepods", "SmCopepods")) %>%
  mutate(Value = as.numeric(Value)) %>% 
  pivot_wider(values_from = "Value", names_from = "Var") %>% 
  mutate(small_large_index =  SmCopepods - LgCopepods)



ggplot(zoo_lg_small, aes(Time, small_large_index)) +
  geom_line() +
  geom_hline(yintercept = 0) +
  facet_grid(EPU~., scales = "free") +
  labs(
    y = "Small-Large Copepod Index\n(More Large Copepods  <----->  More Small Copepods)")

# # This is too many vars
# ecodata::zooplankton_index %>% distinct(Var) %>% pull()
# 
# # Not informative mechanistically
# ecodata::zoo_diversity
# 
# # BOLD move on absolute abundances
# ecodata::zoo_strat_abun

NEEDS: MCC & Lobster Predator Indices

There are two EPU-Scale indices that we need to develop. This is the MCC index, and a lobster predator abundance index.

The Gulf of Maine Coastal Current plays an important role in transporting lobster larva and their recruitment form year-to-year. The degree of “connected-ness” of the Western and Eastern portions of this current have been used in the past to inform expectations of lobster recruitment.

Local/Nearshore Shifts

Conditions closer to the coast show the following long-term trends:

# Temperature
fvcom_trends %>% 
  filter(str_detect(var, "temperature"),
         `Trend?`,
         !area_id %in% c("gom_gbk", "sne")) %>% 
  group_by(area_id) %>% 
  gt::gt() %>% 
  gt::tab_header(
    title = "FVCOM Offshore Long Term Salinity Trends",
    subtitle = "Monotonic Trends Evaluated by Mann-Kendall Test") %>%
  gt::fmt_missing(columns = everything(), missing_text = "")

FVCOM Offshore Long Term Salinity Trends
Monotonic Trends Evaluated by Mann-Kendall Test
var	Trend?	Decadal Rate
eastern maine
bottom_temperature	TRUE	0.007
surface_temperature	TRUE	0.009
new jersey shore
bottom_temperature	TRUE	0.015
surface_temperature	TRUE	0.014
central maine
surface_temperature	TRUE	0.011
eastern mass
surface_temperature	TRUE	0.018
western maine
surface_temperature	TRUE	0.016

# Salinity
fvcom_trends %>% 
  filter(!str_detect(var, "temperature"),
         `Trend?`,
         !area_id %in% c("gom_gbk", "sne")) %>% 
  group_by(area_id) %>% 
  gt::gt() %>% 
  gt::tab_header(
    title = "FVCOM Offshore Long Term Salinity Trends",
    subtitle = "Monotonic Trends Evaluated by Mann-Kendall Test") %>%
  gt::fmt_missing(columns = everything(), missing_text = "")

FVCOM Offshore Long Term Salinity Trends
Monotonic Trends Evaluated by Mann-Kendall Test
var	Trend?	Decadal Rate
central maine
bottom_salinity	TRUE	-0.001
surface_salinity	TRUE	-0.003
eastern mass
bottom_salinity	TRUE	-0.004
surface_salinity	TRUE	-0.006
five fathom bank
bottom_salinity	TRUE	0.007
surface_salinity	TRUE	0.007
long island sound
bottom_salinity	TRUE	0.005
surface_salinity	TRUE	0.006
new jersey shore
bottom_salinity	TRUE	0.004
surface_salinity	TRUE	0.004
southern mass
bottom_salinity	TRUE	-0.003
surface_salinity	TRUE	-0.003
virginia shore
bottom_salinity	TRUE	0.003
surface_salinity	TRUE	0.003
western maine
bottom_salinity	TRUE	-0.002
surface_salinity	TRUE	-0.004

Temperature and Salinity Breaks

# Plot the breaks over the monthly data
ggplot() +
   # Add a vertical line using geom_segment with an arrow
  geom_segment(
    data = filter(
      inshore_tempsal_shifts, 
      str_detect(shift_direction, "Up")),
    linewidth = 0.8,
    aes(x = time, xend = time, y = -Inf, yend = Inf, 
        color = shift_direction),  
    arrow = arrow(length = unit(0.3, "cm"), ends = "last")) +
  geom_segment(
    data = filter(
      inshore_tempsal_shifts, 
      !str_detect(shift_direction, "Up")),
    linewidth = 0.8,
    aes(x = time, xend = time, y = -Inf, yend = Inf, 
        color = shift_direction),  
    arrow = arrow(length = unit(0.3, "cm"), 
                  ends = "first")) +
  geom_line(
    data = filter(inshore_tempsal),
     aes(time, detrended_vals),
    linewidth = 0.4, alpha = 0.5) +
  scale_color_gmri() +
  facet_grid(area_id~Var, labeller = label_wrap_gen(width = 8), scales = "free") +
  scale_x_date(
    breaks = seq.Date(
      from = as.Date("1950-01-01"),
      to = as.Date("2020-01-01") ,
      by = "10 year"),
     limits = as.Date(c("1978-01-01", "2020-01-01")),
    labels = scales::label_date_short()) +
  theme(
    legend.position = "bottom",
    strip.text.y = element_text(angle = 0)) + 
  labs(color = "",
       y = "Measurement",
       x = "Date",
       title = "FVCOM Inshore Scale Sal Metrics",
       subtitle = "STARS Regime Shifts")

Salinity

# Plot the breaks over the monthly data
ggplot() +
   # Add a vertical line using geom_segment with an arrow
  geom_segment(
    data = filter(
      inshore_tempsal_shifts,
      str_detect(Var, "Salinity"),
      str_detect(shift_direction, "Up")),
    linewidth = 0.8,
    aes(x = time, xend = time, y = -Inf, yend = Inf, 
        color = shift_direction),  
    arrow = arrow(length = unit(0.3, "cm"), ends = "last")) +
  geom_segment(
    data = filter(
      inshore_tempsal_shifts, 
      str_detect(Var, "Salinity"),
      !str_detect(shift_direction, "Up")),
    linewidth = 0.8,
    aes(x = time, xend = time, y = -Inf, yend = Inf, 
        color = shift_direction),  
    arrow = arrow(length = unit(0.3, "cm"), 
                  ends = "first")) +
  geom_line(
    data = filter(inshore_tempsal, str_detect(Var, "Salinity")),
     aes(time, detrended_vals),
    linewidth = 0.4, alpha = 0.5) +
  scale_color_gmri() +
  facet_grid(area_id~Var, labeller = label_wrap_gen(width = 8), scales = "free") +
  scale_x_date(
    breaks = seq.Date(
      from = as.Date("1950-01-01"),
      to = as.Date("2020-01-01") ,
      by = "10 year"),
     limits = as.Date(c("1978-01-01", "2020-01-01")),
    labels = scales::label_date_short()) +
  theme(
    legend.position = "bottom",
    strip.text.y = element_text(angle = 0)) + 
  labs(color = "",
       y = "Salinity Anomaly",
       x = "Date",
       title = "FVCOM Hindcast Inshore Salinity",
       subtitle = "STARS Regime Shifts")

inshore_tempsal_shifts %>% 
  filter(str_detect(Var, "Salinity")) %>% 
  group_by(Var) %>% 
  arrange(time, area_id) %>% 
  gt::gt() %>% 
  gt::tab_header(title = "Inshore Salinity Regime Breaks")

Inshore Salinity Regime Breaks
time	area_id	shift_direction
Surface Salinity
1991-07-02	virginia shore	Shift Down
1996-07-02	five fathom bank	Shift Down
Bottom Salinity
1992-08-02	virginia shore	Shift Down
1996-07-02	five fathom bank	Shift Down

# Highlight whichever regions have seen salinity regime change
ggplot() +
  geom_sf(data = filter(
    study_regions, 
    area_id %in% 
      (dplyr::filter(tempsal_shifts, str_detect(Var, "Salinity")) %>% 
      pull(area_id))
    ),
    fill = gmri_cols("gmri blue"), alpha = 0.4) +
  geom_sf(data = new_england) +
  geom_sf(data = canada) +
  
  coord_sf(xlim = c(-78, -66), ylim = c(35.5, 45)) +
  labs(title = "Salinity Regime Change Affected Areas")

Temperatures

# Plot the breaks over the monthly data
ggplot() +
   # Add a vertical line using geom_segment with an arrow
  geom_segment(
    data = filter(
      inshore_tempsal_shifts,
      !str_detect(Var, "Salinity"),
      str_detect(shift_direction, "Up")),
    linewidth = 0.8,
    aes(x = time, xend = time, y = -Inf, yend = Inf, 
        color = shift_direction),  
    arrow = arrow(length = unit(0.3, "cm"), ends = "last")) +
  geom_segment(
    data = filter(
      inshore_tempsal_shifts, 
      !str_detect(Var, "Salinity"),
      !str_detect(shift_direction, "Up")),
    linewidth = 0.8,
    aes(x = time, xend = time, y = -Inf, yend = Inf, 
        color = shift_direction),  
    arrow = arrow(length = unit(0.3, "cm"), 
                  ends = "first")) +
  geom_line(
    data = filter(inshore_tempsal, !str_detect(Var, "Salinity")),
     aes(time, detrended_vals),
    linewidth = 0.4, alpha = 0.5) +
  scale_color_gmri() +
  facet_grid(area_id~Var, labeller = label_wrap_gen(width = 8), scales = "free") +
  scale_x_date(
    breaks = seq.Date(
      from = as.Date("1950-01-01"),
      to = as.Date("2020-01-01") ,
      by = "10 year"),
     limits = as.Date(c("1978-01-01", "2020-01-01")),
    labels = scales::label_date_short()) +
  theme(
    legend.position = "bottom",
    strip.text.y = element_text(angle = 0)) + 
  labs(color = "",
       y = "Temperature Anomaly",
       x = "Date",
       title = "FVCOM Hindcast Inshore Temperature",
       subtitle = "STARS Regime Shifts")

ggplot() +
  geom_sf(data = filter(
    study_regions, 
    area_id %in% 
      (dplyr::filter(inshore_tempsal_shifts, !str_detect(Var, "Salinity")) %>% 
      pull(area_id))
    ),
    fill = gmri_cols("lv orange"), alpha = 0.4) +
  geom_sf(data = new_england) +
  geom_sf(data = canada) +
  
  coord_sf(xlim = c(-78, -66), ylim = c(35.5, 45)) +
  labs(title = "Inshore Temperature Regime Shift Affected Areas")

inshore_tempsal_shifts %>% 
  filter(!str_detect(Var, "Salinity")) %>% 
  group_by(Var) %>% 
  arrange(time, area_id) %>% 
  gt::gt() %>% 
  gt::tab_header(title = "Inshore Scale Temperature Regime Breaks")

Inshore Scale Temperature Regime Breaks
time	area_id	shift_direction
Surface Temperature
1986-05-15	central maine	Shift Down
1986-05-15	western maine	Shift Down
1986-05-15	long island sound	Shift Down
1987-02-15	eastern maine	Shift Down
2002-11-15	eastern mass	Shift Down
2002-11-15	new jersey shore	Shift Down
2003-01-15	western maine	Shift Down
2009-11-15	eastern maine	Shift Up
2009-11-15	central maine	Shift Up
2009-11-15	western maine	Shift Up
2009-11-15	eastern mass	Shift Up
2011-02-15	long island sound	Shift Up
2011-02-15	new jersey shore	Shift Up
2011-02-15	five fathom bank	Shift Up
2011-03-15	southern mass	Shift Up
Bottom Temperature
1986-05-15	long island sound	Shift Down
2002-11-15	southern mass	Shift Down
2002-11-15	new jersey shore	Shift Down
2002-11-15	virginia shore	Shift Down
2009-11-15	southern mass	Shift Up
2011-02-15	long island sound	Shift Up
2011-02-15	new jersey shore	Shift Up
2011-02-15	virginia shore	Shift Up

Days in Key Temperature Ranges

In addition to breaks in absolute temperatures, there is interest in the amount of time spent in favorable (12-18C) and unfavorable conditions (20C).

These use daily bottom temperatures:

# Load the data for the new regions
daily_fvcom_temps <- read_csv(
  str_c(lob_ecol_path, "FVCOM_processed/area_timeseries/new_regions_fvcom_temperatures_daily.csv")
) %>% 
  mutate(
    depth_type = if_else(area_id %in% c("gom_gbk", "SNE"), "offshore", "nearshore"),
    area_id = factor(area_id, levels = areas_northsouth)
  )

# # Monthly Salinity
# monthly_fvcom_sal <- read_csv(
#   str_c(lob_ecol_path, "FVCOM_processed/area_timeseries/new_regions_fvcom_salinity_monthly_gom3.csv")) %>% 
#   mutate(
#     depth_type = if_else(area_id %in% c("gom_gbk", "SNE"), "offshore", "nearshore"),
#     area_id = factor(area_id, levels = areas_northsouth)
#   )




# The degday functions were built with the ability to model daily cycles
# the sine and triangle methods accomodate this
# since we only have daily data the simple average is probably the way to do it
thresh_low <- 10
thresh_up  <- 18

# library(degday)
# Get Monthly Totals in Ranges
dd_monthly <- daily_fvcom_temps %>%
  mutate(
    year = lubridate::year(time),
    month = lubridate::month(time),
    opt_btemp     = if_else(between(bottom_t, 10,18), 1, 0),
    stress_btemp  = if_else(bottom_t > 18, 1, NA),
    cold_btemp     = if_else(bottom_t < 10, 1, NA)) %>% 
  group_by(area_id, year, month) %>% 
  summarise(
    across(
      ends_with("temp"), 
      ~sum(.x, na.rm = T)),
    .groups = "drop") %>% 
  pivot_longer(
    cols = ends_with("temp"), 
    names_to = "var", 
    values_to = "totals") %>% 
  mutate(
    time = as.Date(str_c(year,"-01-01")) + months(month-1),
    area_id = factor(area_id, levels = areas_northsouth),
    var = case_match(
      var,
      "opt_btemp" ~ "Preferred Bottom Temperatures 10-18C", 
      "stress_btemp" ~ "Heat Stress Conditions >18C",
      "cold_btemp" ~ "Below Preferred Conditions <10C")
  )

# Do annual, Monthly values looked insane
dd_annual <- dd_monthly %>% 
  group_by(year, area_id, var) %>% 
  summarise(across(totals, sum)) %>%
  mutate(var_area = str_c(var, area_id, sep = "X"),
         time = as.Date(str_c(year, "-01-01")))

# Plot them
dd_annual %>% 
  ggplot() +
  geom_area(
    aes(year, y = totals, fill = var)) +
  scale_fill_manual(values = c("lightblue",  "#ea4f12", "#057872")) +
  facet_grid(area_id~var) +
  scale_x_continuous(expand = expansion(add = c(0,0))) +
  theme(strip.text.y = element_text(angle = 0),
      legend.position = "bottom") +
  guides(fill = guide_legend(
    nrow = 2,
    title.position = "top",
    title.hjust = 0.5))+
  labs(y = "Days in Range",
       fill = "Daily Temperature Conditions", color = "",
       title = "FVCOM Bottom Temperature Degree-Days")

# # Remove monthly averages, and trends
dd_annual_detrended <-  dd_annual  %>%
  split(.$var_area) %>%
  map_dfr(function(.x){

    # Detrend
    .x <- .x %>%
      arrange(year) %>%
      mutate(yr_num = as.numeric(year))

    # annual trend + monthly average
    trend_mod <- lm(totals ~ yr_num, data = .x)

    # save the results
    .x <- broom::augment(x = trend_mod) %>%
      rename(
        trend_fit = .fitted,
        trend_resid = .resid) %>%
      full_join(.x, join_by(yr_num, totals)) %>%
      mutate(trend_resid = if_else(is.na(totals), NA, trend_resid))

    return(.x)}) %>% 
  mutate(time = as.Date(str_c(year, "-01-01")))



# Plot
dd_annual_detrended %>% 
  filter(area_id %in% tolower(c("GOM_GBK", "SNE"))) %>%
  ggplot(aes(time, trend_resid, fill = var, color = var)) +
  geom_col() +
  geom_smooth(method = "loess", linewidth = 0.6) +
  scale_color_manual(values = c("lightblue",  "#ea4f12", "#057872")) +
  scale_fill_manual(values = c("lightblue",  "#ea4f12", "#057872")) +
  facet_grid(area_id~var, scales = "free", labeller = label_wrap_gen(width= 8)) +
  guides(color = guide_legend(nrow = 3)) +
  labs(title = "EPU Scale Ocean/Climate Metrics - Detrended",
       y = "Departure from Long-Term Trend (days)")

# Run the regime shift test
temp_range_rstars <- dd_annual_detrended %>%
# temp_range_rstars <-   dd_annual %>% 
  split(.$var_area) %>%
  map_dfr(function(.x){
    
    # Seven years
    cutoff_length <- 7
    
    
    # Get the results from that
    x_rstars <- rstars(
      data.timeseries = as.data.frame(
        .x[,c("time", "trend_resid")]),
      l.cutoff = cutoff_length,
      pValue = 0.05,
      Huber = 1,
      Endfunction = T,
      preWhitening = T,
      OLS = F,
      MPK = T,
      IP4 = F,
      SubsampleSize = (cutoff_length + 1)/3,
      returnResults = T) %>% 
      mutate(
        Value = .x$totals,
        shift_direction = case_when(
          RSI > 0 ~ "Shift Up", 
          RSI < 0 ~ "Shift Down",
          TRUE ~ NA))
    
    },
    .id = "Var"
  )%>% 
  separate(Var, into = c("Var", "area_id"), sep = "X") %>% 
  mutate(area_id = factor(area_id, levels = areas_northsouth))

Temperature Suitability Shifts

The results can be seen below:

# Summarise the breakpoint locations
temp_suit_shift_points <- temp_range_rstars %>% 
  filter(RSI != 0) %>% 
  dplyr::select(time, Var, area_id, shift_direction)


# Plot the breaks over the monthly data
ggplot() +
  # Add a vertical line using geom_segment with an arrow
  geom_segment(
    data = filter(
      temp_suit_shift_points, 
      str_detect(shift_direction, "Up")),
    linewidth = 0.8,
    aes(x = time, xend = time, y = -Inf, , yend = Inf, 
        color = shift_direction),  
    arrow = arrow(length = unit(0.3, "cm"), ends = "last")) +
  geom_segment(
    data = filter(
      temp_suit_shift_points, 
      !str_detect(shift_direction, "Up")),
    linewidth = 0.8,
    aes(x = time, xend = time, y = -Inf, , yend = Inf, 
        color = shift_direction),  
    arrow = arrow(length = unit(0.3, "cm"), ends = "first")) +
  geom_line(
    data = temp_range_rstars,
     aes(time, Value),
    linewidth = 0.4, alpha = 0.5) +
  scale_color_gmri() +
  facet_grid(area_id ~ Var, labeller = label_wrap_gen(width = 8), scales = "free") +
  scale_x_date(
    breaks = seq.Date(
      from = as.Date("1950-01-01"),
      to = as.Date("2020-01-01") ,
      by = "10 year"),
     limits = as.Date(c("1970-01-01", "2020-01-01")),
    labels = scales::label_date_short()) +
  theme(
    legend.position = "bottom",
    strip.text.y = element_text(angle = 0)) + 
  labs(color = "",
       y = "Measurement",
       x = "Date",
       title = "Lobster Thermal Preferences - Detrended",
       subtitle = "STARS Regime Shifts")

Based on the annual totals, we see limited breakpoints in suitable thermal habitat.

temp_suit_shift_points %>% 
  group_by(Var) %>% 
  arrange(time, area_id) %>% 
  gt::gt() %>% 
  gt::tab_header(title = "Temperature Suitability Scale Breaks")

Temperature Suitability Scale Breaks
time	area_id	shift_direction
Heat Stress Conditions >18C
1992-01-01	virginia shore	Shift Down
2004-01-01	rhode island shore	Shift Down
2011-01-01	five fathom bank	Shift Up
2012-01-01	rhode island shore	Shift Up
Below Preferred Conditions <10C
2011-01-01	new jersey shore	Shift Down

And restricted to these areas

ggplot() +
  geom_sf(data = filter(
    study_regions, 
    area_id %in% temp_suit_shift_points$area_id),
    fill = gmri_cols("lv orange"), alpha = 0.4) +
  geom_sf(data = new_england) +
  geom_sf(data = canada) +
  
  coord_sf(xlim = c(-78, -66), ylim = c(35.5, 45)) +
  labs(title = "Affected Areas")

Summary Figures / Tables

Do the trend evaluation for everything:

Summary table should have the region, the variable, whether there is a trend or not, what the rate is, then whether there have been breakpoints, and when they were (mm/yyyy).